Dynamic Analysis 17-654/17-754: Analysis of Software Artifacts - - PDF document

1

• Dynamic Analysis

17-654/17-754: Analysis of Software Artifacts Jonathan Aldrich

• Part 1: Performance Analysis
• Your boss tells you, “make it run faster.”

What do you do?

2

• Speedup Analysis Scenario
• Assumptions
• Procedure x takes 75% of execution time
• We can speed up x by a factor of 3
• How much faster will the program run?
• Amdahl’s Law
• P = % of program you can speed up
• S = speedup of that part of the program
• Maximum overall speedup:

1 (1-P) + P/S

time to run sped-up program part time to run rest of program

3

• 80/20 rule (Pareto principle)
• 20% of the program takes 80% of the

time

• No point in optimizing anything until you

know what 20% is causing the problem!

• Another way of stating Amdahl’s law
• Getting Good Performance
• What matters?
• Algorithms
• Parallelism
• Cache

performance

• Caching
• trade space for time
• Doing less work
• trade precision for

time

• What doesn’t?
• Inlining vs. calls
• field reads vs.

accessors

• Recursion vs.

loops

• Low-level

arithmetic

What’s the cost of optimization?